NuclePhaser
A Napari plugin to detect and count nuclei on phase contrast images
A Napari plugin to detect and count nuclei on phase contrast images
napari-nuclephaser utilizes Ultralytics YOLO object detection models and obss/sahi sliced inference methods to detect cell nuclei on phase contrast (and other brightfield) images of any size, including large whole slide ones.
We trained a series of YOLOv5 and YOLOv11 models to detect nuclei on phase contrast images. It can be used for counting cells or for individual cell tracking (using nuclei detections as tracking marks). Prominent features of this approach are:
- Napari-nuclephaser plugin inclues obss/sahi functionality, allowing detection on images of arbitrary sizes.
- YOLO models are fast, providing reasonable inference speed even with CPU.
- Ability to predict and automatically count nuclei on stacks of images, making it convenient for cell population growth studies and individual cell tracking.
Result of object detection model inference is highly dependent on confidence threshold parameter.
We created several calibration (finding optimal confidence threshold) algorithms that allow adjusting models to specific use cases (cell types, magnifications, illumination settings, cameras etc.):
- Calibration using known number of objects on an image. Doesn't produce accuracy metrics.
- Calibration using fluorescent nuclei stain image (for example, DAPI image). Produces accuracy metrics.
- Calibration using manual annotation of nuclei. Produces accuracy metrics.
Apart from optimal confidence threshold search, these algorithms return accuracy metrics for specific use cases. Given that the calibration image is large, only part of it is used for search of threshold, while the second part is used for evaluation model's accuracy. Accuracy metrics are Mean Absolute Percentage Error (MAPE) and prediction-ground truth scatterplot, which shows how well model performs with different densities of cells.
Currently only YOLOv5n, YOLOv5s, YOLOv11n and YOLOv11s models, as well as fluorescent nuclei detector YOLOv5n are available (downloaded automatically with pip install napari-nuclephaser). We are currently working on adding YOLOv5m-x and YOLOv11m-x models.
[!NOTE] Feel free to use the models published there without the plugin!
napari-nuclephaser plugin offers following widgets:
- Widget for inference on single image. Result can be in the form of points or boxes with or without confidence scores. Automatically returns number of cells in the name of result layer.
- Widget for inference on stack of images. Optionally can create .csv or .xlsx file at given location with counting results.
- Widget for calibration using known number of cells.
- Widget for calibration using fluorescent nuclei image (fluorescent nuclei detection model is used as a perfect predictor).
- Widget for calibration using manual annotations.
- Widget for transforming Napari Points layer into Labels layer, which allows turning detection in tracking algorithms-digestible form (in particular, btrack).
- Widget for counting number of points in Points layer.
Hover with a mouse over a parameter to get a tooltip with short description of functionality. We are currently working on documentation with full description of widgets and their parameters. For now, if you have questions about widgets and/or parameters, please refer to ultralytics documentation and obss/sahi documentation
We are currently working on a paper with full description of our approach and how we trained and tested our models.
This napari plugin was generated with copier using the napari-plugin-template.
Option 1: Using Anaconda (recommended)¶
We recommend installation using Anaconda Distribution
-
Install Anaconda with Installation instructions
-
Open Anaconda Prompt using Search Bar or Anaconda Navigator
-
Create new environment with default anaconda packages using command
conda create --name napari-env anaconda
- Activate new environment using command
conda activate napari-env
- Install Napari using command
pip install napari[all]
- Verify napari installation using following command. It should open napari GUI.
napari
- Install napari-nuclephaser plugin using command
pip install napari-nuclephaser
- Plugin is ready to be used! Start napari by typing
napari
Initialize plugin's widgets by opening Plugins window and choosing NuclePhaser.
Installation with GPU¶
If you have NVIDIA GPU with CUDA, you can significantly increase plugin's speed.
To install GPU-powered version of the plugin, you first need to do all the steps for the installation using Anaconda (above). Then you need to:
- Install CUDA using official instructions
[!NOTE] Check which versions of CUDA are supported by current torch installation and consider installing earlier ones
- Check CUDA installation with nvidia-smi command in the command line.
nvidia-smi
- In the environment with napari and napari-nuclephaser installed, install CUDA-supported torch by typing specific command for your system, which can be found at torch installation page. For example, if you have Windows-based system and CUDA 12.6, your line should look like
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
[!WARNING] During our tests, torchvision wasn't installed using this line. To avoid that, add -U after install:
pip3 install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
Option 2: Using standalone napari app (simpler)¶
-
Download and install napari as standalone app using installation instructions
-
Search, download and install napari-nuclephaser plugin by opening the app, navigating to Plugins window and choosing Install/Uninstall plugins...
Contributing¶
Contributions are very welcome. Tests can be run with tox, please ensure the coverage at least stays the same before you submit a pull request.
License¶
Distributed under the terms of the MIT license, "napari-nuclephaser" is free and open source software
Issues¶
If you encounter any problems, please file an issue along with a detailed description.
Supported data:
- Information not submitted
Plugin type:
GitHub activity:
- Stars: 1
- Forks: 0
- Issues + PRs: 0